The deployment of Multipath QUIC (MPQUIC) in Unmanned Aerial Vehicle (UAV)-assisted Space-Air-Ground Integrated Networks (SAGINs) is severely hampered by the out-of-order (OFO) packet delivery problem. Frequent stream handovers, high mobility, and massive multi-access contention in these networks introduce severe transport-layer challenges. Existing solutions typically isolate multipath scheduling from congestion control, which leads to suboptimal performance and transient congestion in highly dynamic environments. To overcome these limitations, this paper proposes the GPR Hierarchical Synergistic Framework, representing the first joint optimization of multipath scheduling and congestion control for multi-access MPQUIC in SAGINs. Our framework introduces the GradNorm Probabilistic Self-Predictive (GPASP) module to forecast latent states and filter task-irrelevant information in high-dimensional, noisy observation spaces. Furthermore, we develop a Proactive Handover-Aware Congestion Control (PHACC) algorithm that leverages neural network-driven decisions to proactively distinguish handover-induced packet losses from actual network congestion. To address decision-making lag caused by neural network inference latency, a Neural-network Preference Estimation (NNPE) algorithm is designed for highly efficient, real-time scheduling. Extensive ns-3 simulations demonstrate that the proposed framework significantly outperforms state-of-the-art baselines, achieving substantial goodput improvements and a marked reduction in OFO degrees.

翻译：在无人机辅助的空天地一体化网络中部署多路径QUIC协议时，乱序数据包交付问题严重制约了其性能。此类网络中频繁的流切换、高移动性以及大规模多接入竞争给传输层带来了严峻挑战。现有解决方案通常将多路径调度与拥塞控制机制相互隔离，导致在高度动态环境中性能欠佳并易引发瞬时拥塞。为克服这些局限性，本文提出了GPR分层协同框架，首次实现了空天地一体化网络中多接入MPQUIC的多路径调度与拥塞控制联合优化。该框架引入梯度范数概率自预测模块，用于在高维噪声观测空间中预测潜在状态并过滤任务无关信息。此外，我们开发了主动切换感知拥塞控制算法，该算法利用神经网络驱动的决策机制，主动区分因切换引发的数据包丢失与实际网络拥塞。针对神经网络推理延迟导致的决策滞后问题，设计了神经网络偏好估计算法以实现高效实时调度。基于ns-3的大规模仿真实验表明，所提框架显著优于现有先进基线方案，在实现吞吐量大幅提升的同时，有效降低了乱序程度。